Trustworthy decentralized knowledge graph

Abstract

A knowledge graph (KG) is a structured representation of information that connects entities and their relationships in a graph format. It is a powerful tool for organizing and querying vast amounts of data, enabling more efficient data retrieval and analysis. In the context of Web 3.0, decentralization plays a crucial role in reshaping how information is managed and shared online. Decentralized Knowledge Graphics (DKG) represent a paradigm shift in knowledge management, where the creation, transferring, and query of KG data are distributed among a network of participants. This decentralized approach promotes collaboration, diversity of perspectives, and collective intelligence, leading to more dynamic and inclusive knowledge ecosystems. However, the openness and decentralization of these networks can also lead to security concerns, including data poisoning attack, leakage of sensitive information and malicious data manipulation. In this thesis, we analyze various security risks in DKG and propose novel solutions to build and manage trustworthy DKG. First, we focus on DKG construction process, i.e., security issues in Federated Knowledge Graph Embedding (FKGE). FKGE is an emerging collaborative learning technique to construct DKGs. However, poisoning attacks in FKGE, which lead to biased decisions by downstream applications, remain unexplored. We systematize the risks of FKGE poisoning attacks, from which we develop a novel framework for poisoning attacks that force the victim client to predict specific false facts. Based on the attack framework, we explore a blockchain-based defense strategy that can solve this problem by changing the aggregation paradigm. Second, we consider the data ownership protection in DKG sharing. Without central oversight, it is challenging to enforce restrictions or audit data access and usage effectively across all DKG participants. We propose PISTIS, the first DKG provides SPARQL queries with data ownership guarantees. Third, we consider the data integrity and query verifiability requirements in DKG queries. While existing studies focus on ensuring data integrity, how to ensure query verifiability - thus guarding against incorrect, incomplete, or outdated query results - remains unsolved. We propose VERIDKG, the first SPARQL query engine for DKG that offers both data integrity and query verifiability guarantees. In summary, the thesis aims to build secure and trustworthy decentralized knowledge graphs from the perspective of construction, sharing and query. Theoretical analysis and experimental evaluations demonstrate the performance advantages of our methods with provable security.